ACC Reward Location Information Is Carried by Hippocampal Theta Synchrony and Suppressed in a Type 2 Diabetes Model

Abstract

The anterior cingulate cortex (ACC) is important for higher-order cognitive functions, emotional responses, and monitoring internal states. ACC dysfunction has been implicated in an array of psychiatric and neurodegenerative disorders which have a bidirectional relationship with the metabolic disorder Type 2 diabetes (T2D). T2D is a chronic disease characterized by hyperglycemia, loss of insulin signaling, neuroinflammation, and increased morbidity and mortality chances. To better understand the functional effects of T2D on ACC information processing, we delivered an intermittent, low-dose streptozotocin (STZ) protocol to rats (all male due to female insensitivity to STZ) which led to lasting hyperglycemia and recorded single neurons during a delayed alternation task. We observed changes in spatial and reward processing in spite of no differences in overall behavioral accuracy, though we did find hyperglycemic animals spent less time at the reward site. Hyperglycemic animal (n = 5) ACC neurons had higher spatial information scores and changes in the allotment of spatial coding assets. Specifically, the hyperglycemic group had greatest spatial information during the reward approach, while in controls (n = 3) it was uniformly distributed. We found that state space separation and decoding accuracy were greater in control ensembles at the reward location. Furthermore, control hippocampal theta phase-locked cells had the strongest reward coding, and this effect was absent in hyperglycemic animals, leading to a muted reward location representation, despite increased reward approach coding. T2D inferred a nuanced and layered effect on ACC activity, leading to reward coding deficits, a reduced post-reinforcement pause, and a differential change in spatial coding properties.

Significance Statement

Type 2 diabetes (T2D) is a major health challenge in the twenty-first century and makes patients more prone to psychiatric and neurodegenerative disorders. In this paper, we show that neural information processing in the anterior cingulate cortex (ACC), a central area for goal-directed behavior, is altered in multiple ways in a T2D rodent model during a spatial working memory task. Notably, data reveal altered spatial information in ACC cells and muted reward location coding specifically in hippocampal theta-modulated ACC neurons and ensembles. The finding of altered reward processing in the ACC that manifested as a shorter postreinforcement pause invites profound questions given the importance of lifestyle interventions to treat T2D.

Introduction

The increasing prevalence of diabetes in the human population has raised several alarms in the scientific research and health community, and many think it to be a leading global epidemic of the twenty-first century. Globally, 1 out of every 10 people have diabetes (Saeedi et al., 2019), 90% of whom have Type 2 diabetes (T2D; Lundqvist et al., 2019; Saeedi et al., 2019). T2D is a chronic disease linked to obesity, which often has morbid consequences. T2D is characterized by hyperglycemia and insufficient insulin secretion or resistance leading to chronic damage to nerves, blood vessels, and the surrounding tissues and organs (Reijmer et al., 2013; Geijselaers et al., 2015). Impaired functionality of several organs, including multiple brain areas, leads to impaired cognitive functions and mild cognitive impairment (MCI; Groeneveld et al., 2018) and subsequently may lead to dementia and even Alzheimer's disease (AD).

Working memory is a multimodal cognitive process that involves brain areas from the prefrontal cortex (PFC) and the medial temporal lobe. The anterior cingulate cortex (ACC), an integral part of the medial PFC, is especially important for integration of diverse context and task-relevant information to enable for spatial navigation, outcome monitoring, decision-making, and working memory (Hyman et al., 2012; Heilbronner and Hayden, 2016). The ACC is also involved in the reward processing circuitry, hedonic/pleasure processing, and regulation of emotions, making it an integral part of many psychopathologies and mood disorders, especially depressive mood states.

Research has consistently shown that the ACC plays a key role in the processing of task-relevant information. Relatedly, ACC neurons respond to behavioral sequences, task phase, temporal information, and rules (Durstewitz et al., 2010; Hyman et al., 2013; Ma et al., 2014; Wirt et al., 2024). The ACC has a rich network of connections, both direct and indirect, to communicate with the hippocampus (HPC) that are important for spatial navigation, working memory, and reward-based learning (Buzsáki and Moser, 2013; Fillinger et al., 2017, 2018). Both ACC and HPC neurons exhibit spatial-related firing over the entire environment, which is heightened around reward locations. In humans, altered reward responses have been observed in multiple disorders including depression with patients exhibiting anhedonia (Carter and Swardfager, 2016), substance abuse patients with increased ACC and HPC responses to drug cues (Fotros et al., 2013; Koob and Volkow, 2016), and obese individuals with heightened HPC and ACC activation when viewing sugary foods (Martin et al., 2010).

Multiple studies have indicated that people with T2D are at risk of developing cognitive impairments (Strachan et al., 1997, 2003) and impairments in information processing related to working memory tasks, as well as some aspects of attention and key mood states (Sommerfield et al., 2004). The prevalence of MCIs in T2D patients is estimated to be as high as 45% (You et al., 2021). Additionally, T2D is a leading risk factor for developing AD and other neurodegenerative disorders, with rates 65% higher in diabetic individuals (Arvanitakis et al., 2004). While the link between T2D and AD is poorly understood, the presence of common cognitive impairments between T2D and prodromal AD suggests that both diseases are affecting similar parts of the brain. There is very little known about why T2D leads to such effects, but establishing that T2D pathologies alter cognitive processing would serve as an important step to bridge this gap. To investigate T2D's pathological effects on cognition, we utilized a well characterized cognitive task, spatial delayed alternation, and examined effects on spatial and reward processing. The delayed alternation task is known to be affected by T2D (Anthony et al., 2006; Frank et al., 2016), and likewise, we concentrated on two brain areas important for spatial working memory that are affected by T2D, the ACC (Martin et al., 2010) and the HPC (Monereo-Sánchez et al., 2023). Using a low dose of streptozotocin (STZ) protocol to induce chronic hyperglycemia, a key pathology of T2D, we recorded neuronal ensembles in the ACC and HPC, while animals performed a T-maze alternation task with variable length delays. This protocol has been shown to reliably create lasting hyperglycemia while leading to slight long-term memory deficits, minimal weight loss, and increased tau phosphorylation in the HPC (Murtishaw et al., 2018; Wirt et al., 2021).

Materials and Methods

Subjects

Subjects were eight male Long–Evans rats (8–12 months) obtained from Charles River Laboratories. Rats were on a restricted food intake of 25 g per day while performing the behavioral task. The rats were housed singly and kept on a 12 h light/dark cycle. Training and recording sessions were carried out during the light cycle and were recorded from 3 d per week. All experimental procedures were approved by the University of Nevada Las Vegas Institutional Animal Care and Use Committee.

Apparatus

We used a custom-built T–maze with three movable doors surrounding a delay area. It was made using white, 14.5-cm-wide textured plastic floors with 28-cm-high walls made from white corrugated plastic. The central stem of the T was 63.5 cm long connecting to the choice arms at the top of the stem reaching 162.5 cm. Each choice arm was roughly 74 cm in length with a small circular reward well near the end flush with the floor. Sugar-free chocolate beverage was delivered to the reward wells through a small tube controlled by a photobeam tripped solenoid valve. The maze was elevated from the ground and surrounded by black curtains with distinctive visual cues attached. The maze and cues remained in the same location during the entirety of the experiment.

Behavioral training

Prior to surgery and injections, rats were trained in multiple phases to perform a delayed spatial alternation task on a T-maze. Rats were first exposed to the maze to habituate them to a new environment. Reward wells were baited with a sugary confectionate treat (Froot Loops; Kelloggs) to encourage exploration. Once the rats were acclimated, rats began the first phase of training; it consisted of forced continuous alternating trials. Rats were placed in the sequester area of the maze with all doors closed. The center stem door opened, and rats would travel to the end of the stem, where only one choice arm would be open on each trail. To encourage the return, the area was initially baited every trial and then periodically after returning became habitual. After sufficiently training the rats in the forced alternation task (3 ± 2 sessions), both arms were opened for the second phase allowing rats to ad libitum choose the correct arm. In the third phase, rats were introduced to the sugar-free chocolate beverage as a reward.

For all subsequent training and testing sessions, rats were placed in the enclosed sequester area with no choice arm barriers and allowed to run for 30–50 trials. The experimenter stayed outside the curtained enclosure for the duration of the session. Animal behavior was observed using a video monitor connected to a tracking system. On each trial with correct arm choice, a drop of sugar-free chocolate beverage was delivered to the well in the arm following arm entry. Incorrect trials resulted in the end of a trial without a reward, and the rat was required to return to the sequestered area on the correct return arm.

Surgery and electrophysiology

Rats were anesthetized using isoflurane gas (1–3%) and surgically implanted with a 32-moveable tetrode hyperdrive affixed to the animal's skull. Tetrodes were made from a single 13 μm wire folded and wound to produce four independent channels. Sixteen of the tetrodes were targeted bilaterally to the ACC (2.5 mm anterior; +0.5 lateral; eight left and eight right) and 16 bilaterally to dorsal CA1 (3.5 mm posterior to the bregma: +2.5 lateral; eight left and eight right). Two stainless steel screws were placed posterior, just above the cerebellum, used as grounding wires and soldered into the electrode interface board (EIB; Plexon), typically done in rodent in vivo recordings, once tetrodes were positioned directly over the target brain areas. The implant was fixed in place using dental acrylic on top of the skull. After the dental acrylic hardened, tetrodes were lowered 400 mm into the cortex. Following a 7 d recovery period, tetrodes were then slowly lowered ventrally into the ACC (∼2.5 mm, 10° angle) and the pyramidal cell layer of dorsal CA1 using known electrophysiological markers. Each electrodes was connected to a 128-channel EIB that could plug into four separate head stages (Intan Technologies). Electrophysiological signals were digitized and sent through two tethered cables into a RHD 2000 USB interface board (Intan Technologies), which is made visible through the Open Ephys interface. Data were sampled at a rate of 30 KHz. Continuous data were passed through a bandpass filter (0.1–6 KHz).

Treatment group

To induce a chronic hyperglycemic state, we used a modified STZ protocol, which consisted of staggered low doses of STZ injections. Our original goal was to achieve chronic levels of hyperglycemia with little changes in body weight. We used a lower dose of STZ due to the additional stress of neural implant surgery. We used male Long–Evans due to female rodents being resistant to STZ treatment (Leiter, 1982). Female mice have historically been resistant to STZ treatment due to the protective aspects of the X chromosome containing most of the immune-related genes as well as sex differences in hormones, with estrogen having a protective effect on pancreatic β-cells (Liu and Mauvais-Jarvis, 2010; Migliore et al., 2021). Female mice become more susceptible to STZ treatment after menopause is induced (Keck et al., 2007). STZ (Sigma-Aldrich) was prepared fresh prior to administration by dissolving the drug in 0.1 M sodium citrate buffer, pH 4.5, for a final concentration of 20 mg/kg/ml. Following a 6 h fast, animals were injected with STZ via intraperitoneal injection, pH 4.5, at final concentration of 20 mg/kg/ml. Our protocol ensured that a sustained hyperglycemic state was based on achieving a group average fasting blood glucose level of ≥250 mg/dl consistent with DM levels. The modified protocol of staggered STZ injections on Days 1, 2, 3, 14, and 15, followed by supplemental injections of STZ on Days 35 and 36, achieved the criteria for sustained hyperglycemia.

Neural data analysis

LFP analysis

All analyses were performed in MATLAB (MathWorks) using custom-written scripts. Local field potential (LFP) signals were analyzed from leads in each area and each hemisphere (16 in total). These leads were selected based on visual inspection in order to minimize noise. Only one wire from each tetrode could be selected. LFP signals were first notch filtered to remove 60 Hz noise.

Analysis of unit data

After each recording session, spike data were read into a computer work station and translated into (.nex) format using a custom-written MATLAB code. Single-unit data were sorted based on waveform characteristics using Plexon Offline Sorter (Plexon). Spiking data were binned into 200 ms windows and Gaussian smoothed to create firing rate time series. Spatial information was created by binning the T-maze into 12 bins; then we examined neural firing rates that corresponded to the animal's location on the maze to create a spatial information score for each cell (using standard methods; Skaggs et al., 1992).

Spatial information

For each cell, the spatial information score was calculate from recordings in the T-maze as follows:

$$\mathrm{Spatial}\mathrm{i}\mathrm{nformation}\mathrm{s}\mathrm{core}=\sum _{i}pi{\displaystyle \frac{{\lambda }_i}{\lambda }\mathrm{Lo}{\mathrm{g}}_2{\displaystyle \frac{{\lambda }_i}{\lambda },}}$$

where ${\lambda }_i$ is the mean firing rate of a unit in the ith bin, λ is the overall mean firing rate, and $pi$ is the probability of the animal being in the ith bin. Cells were classified as place cells by comparing with bootstrapped time shuffles to create an empirical distribution (p < 0.01). Spatial information was calculated separately for left and right trials, and time-shuffled bootstrap distributions were created for each trial type. A cell was deemed spatially tuned if the spatial information on at least one trial type was in the top 1% of the empirical distribution.

Left/right trial selectivity

To examine left/right selectivity for ACC neurons between control and hyperglycemic animals, we calculated left/right selectivity indices for each maze location. We achieved this by taking all the trials for the left or right, and we examined firing rates for each maze location. We then performed a two-factor ANOVA (group × maze location) to examine selectivity bias as follows:

$$d_i^{\mathrm{\prime }}={\displaystyle \frac{⟨\{r_i(t)|t\epsilon A\}⟩-⟨\{r_i(t)|t\epsilon B\}⟩}{\sqrt{{\sigma }_{i,t\epsilon A}^2+}{\sigma }_{i,t\epsilon B}^2}.}$$

Mahalanobis distance

To examine population dynamics and how ensemble activity states differed between control and hyperglycemic animals, we calculated the Mahalanobis distance (dMah) using a custom-written MATLAB code (Durstewitz et al., 2010; Hyman et al., 2012). We accomplished this by randomly selecting seven correct trials for both left and right trials. Each of the 14 trials was split into firing rates for the 12 bins of the maze. We randomly selected 15 neurons, leaving us with a 168 × 15 matrix. To examine population dynamics in lower-dimensional space, we calculated principal component analysis (PCA) on the subset using the MATLAB function pca using a custom-written MATLAB code. We examined the top three principal components (PCs) at each maze location with dMah to see differences in left/right trials.

General linear model–Bonferroni’s correction

We used the generalized linear model (GLM) to determine firing rate variation for individual cells in relation to different task epochs between left and right trials on the maze. We created a series of models that had contrasting firing rate changes at each location for left and right trials. First a model with an increase at Position 1 on left trials and decrease at Position 1 on right trials, with all other locations flat, then a model for Position 2, and so on. Each model was designed to detect maze position specific differences between left and right trials. We used the fitglm function in MATLAB. We used the tstat output of the model to assess the significance of the model and then Bonferroni-corrected for multiple comparisons (p < 0.05).

Linear decoder

We performed a leave-one-out decoding analysis based on linear discriminant functions to decode the accuracies of ACC ensembles over groups across the entire maze. The decoding procedure used the values from the dMah from each bootstrapped ensemble. Linear discriminant analysis finds a (hyper)plane optimally separating two clouds of data points by maximizing the difference between the class means while minimizing within-class covariances. The classifier works by determining how many of the dMah values from the test set fall onto the wrong side of the separating hyperplane, attributing the values to a wrong trial. Optimal predictions were evaluated using a bootstrap analysis. It accomplishes this by randomly shuffling ensembles in blocks 1,000 times. Each of the blocks corresponded to maze location and left or right trials.

Phase locking

For each cell, we calculated the degree of theta and nontheta entrainment for the ACC and HC throughout the whole maze and looked at the maximum value from all comparisons. To analyze phasic modulation of unit spike trains, we first Hilbert-transformed the theta-ranged LFPs to extract the instantaneous phase using the MATLAB function hilbert. We found the phase for each spike during all trials in the session as follows:

$$\mathrm{PLV}=\left|{\displaystyle \frac{1}{N}\sum _{j=1}^N\exp (i[{\varphi }_a(j\mathrm{\Delta }t)-{\varphi }_b(j\mathrm{\Delta }t)])}\right|.$$

The phase of two signals was extracted by the Hilbert transformation and signed as ${\varphi }_a$ and ${\varphi }_b$. The phase-locking value (PLV) was calculated as follows: N stands for the length of time series and 1/Δt is the sampling frequency. The PLVs have values with [0 1]; here 0 represents no phase synchrony, and 1 is the full-phase synchrony. We had 16 channels of LFP data equally split bilaterally targeting HPC and ACC. The PLV was calculated for each pair of channels for left and right trials separately. Then, for each cell, we selected the maximum PLV across both ACC and HPC channels. To determine statistical significance, for each cell and the selected ACC and HPC LFP, 1,000 times shuffled bootstraps were used to create an empirical distribution (p < 0.05).

Results

STZ animals had blood glucose levels above clinical threshold for T2D in humans

Following the series of STZ injections, animals in the experimental group exhibited a sustained fasting blood glucose reading >250 mg/dl (Fig. 1D). The threshold of ≥250 mg/dl was selected based on clinical criteria to establish a chronic hyperglycemic state consistent with T2D. The experimental group mean was 358.1 ± 36.7 mg/dl prior to the first recording session. A two-factor ANOVA revealed significant main effects for group (F_(1,33) = 49.4; p = 1.2 × 10⁻⁷), day (F_(3,33) = 13.05; p = 1.63 × 10⁻⁵), and a significant interaction (F_(3,33) = 14.06; p = 8.9 × 10⁻⁶; Fig. 1D). These results align with our previously published work with this same protocol (Murtishaw et al., 2018; Wirt et al., 2021), showing that the STZ protocol lead to a lasting hyperglycemic state.

Figure 1.

Delayed alternation task and hyperglycemia. A, The modified figure 8 T-maze used for the variable length delayed alternation task. The maze was divided into 12 distinct zones, based on different cognitive meanings to the rats: the delay box (Zone 1), decision point (Zone 4), reward (Zone 7), and return arms (Zones 8–12). B, Representative image of a Long–Evans rat, implanted with a 32 tetrode microdrive, with 16 tetrodes aimed bilaterally at ACC and HPC (CA1), respectively. C, Histology. Representative examples of 40-μm-thick Nissl-stained coronal sections from the HPC (CA1) and ACC showing clear tetrode tracks in both areas. D, Blood glucose values from all eight animals. Blood glucose index values are shown on the y-axis, and the date of testing is on the x-axis. The dashed line indicates threshold for hyperglycemia in humans. E, Animal weights relative to the baseline. On the y-axis are percentage changes in weights relative to before the first injection for the 8 weeks following the start of the injection protocol. Surgeries took place between Weeks 2 and 3. Control animals are in blue, and STZ are in pink. F, Behavioral accuracy between the two groups remained comparable. Proportion of correct trials on y-axis. G, The total number of trials during each session. No difference was found between the number of completed trials between groups. H, Average trial duration. On the y-axis is the mean time between the start of one trial and the start of the next. This includes the delay period, the response, reward consumption, and return time. No differences were found between groups. I, Average moving speeds per trial. Movement speed in cm/s is shown on the y-axis. No differences were found between groups. J,K, Dwell time by maze location. On the y-axis is mean time spent in maze locations (x-axis) in seconds. STZ animals spent significantly less time at the reward zone on rewarded trials (J), while there was no difference in dwell time at any location on error trials (K).

We also analyzed changes in body weight for all animals. While we did find significant main effects for group (F_(1,61) = 6.93; p = 0.01) and day (F_(7,61) = 7.44; p = 5.6 × 10⁻⁶), we did not find any significant interaction (F_(7,61) = 1.28; p = 0.28; Fig. 1E). Most of the group differences occurred in the weeks following surgery, perhaps reflective of the increased metabolic wear and tear of the hyperglycemic state, which could slow surgical recovery while increasing inflammation. Notably, all recordings took place 5.5 weeks or longer after surgery, a time at which there were no differences between groups (p > 0.05).

No differences in overall working memory performance

We have previously reported that STZ-treated animals were impaired on long-delay (>20 s) trials during delayed spatial alternation (Wirt et al., 2021), so for the current analysis we concentrated on short-delay trials. We found no difference between groups in session accuracy for short-delay trials (F_(1,33) = 0.0012; p > 0.05; Fig. 1E). We should note that for all electrophysiological analysis in this paper, only activity from correct trials was used. We also analyzed the total number of trials completed per session to determine if motivation was different between groups. We found no difference in total trials per session (F_(1,33) = 0.2; p = 0.66; Fig. 1F). Additionally, we investigated whether there was a difference in the duration of trials between groups, which could suggest that motivation differed since more trials in a shorter time period could be considered as more motivation. However, we found no difference between groups in mean trial duration (F_(1,33) = 0.23; p = 0.63; Fig. 1G). Lastly, we examined average running speed over the course of the entire maze, and we found no difference between STZ and control groups (F_(1,33) = 3.16; p = 0.085). These behavioral analyses showed that there were no differences in gross behavioral measures (accuracy, trial length, number of trials, average speed) during short-delay trials, revealing that chronic hyperglycemia left short-delay working memory behavior intact.

Hyperglycemic animals did not linger at reward ports like control animals

While our analysis of gross behaviors did not find any group differences, it remained possible that subtle behavioral differences did not manifest into full on gross behavioral changes. To examine more closely, we looked at the time spent in each maze location. To examine this, we ran a two-factor ANOVA (group × location) on dwell time at each location. We found a significant main effect for location (F_(19,640) = 69.06; p > 0.001), but no main effect for the group (F_(1,640) = 0.62; p = 0.433). Importantly, the interaction term was significant (F_(19,640) = 5.4; p > 0.001), and follow-up tests revealed group differences at two locations (Fig. 1J). At the reward location, control animals spent more time than hyperglycemic animals, nearly twice as long. Since experimenter observation and video evidence suggested that hyperglycemic animals were consuming the chocolate milk reward, the short duration at the reward site could indicate quicker reward consumption and a lack of any sort of postreward pause. To ensure that this behavioral difference was related to reward consumption and reward expectation or feedback, we also examined dwell time for error trials (Fig. 1K). We found no interaction between group and location (F_(19,640) = 1.33; p = 0.16). This showed that control animals were pausing after receiving reward, while hyperglycemic animals were barely stopping their runs to consume. Since we found no difference in overall trial time between groups, there must have been other maze locations where hyperglycemic animals lingered more than controls. Indeed, the other location we found with a difference was the last turn before returning to the sequester zone and here hyperglycemic animals dwelled longer. Overall, these results show that reward-related behavior did vary between groups but in a subtle way that did not affect overall task performance or efficiency.

Increased ACC spatial information and prereward coding in hyperglycemic animals

During 34 sessions from 8 rats (control, n = 3; STZ, n = 5), we recorded a total of 569 ACC cells (control, n = 216; STZ, n = 353). We first looked for gross overall differences in ACC single units and found no significant change in overall mean firing rates associated with chronic hyperglycemia (F_(1,568) = 2.107; p > 0.05; Fig. 2B). We next examined spatial information scores in these populations and found that neurons from the STZ group had higher mean spatial information than found in the control group (F_(1,568) = 15.261; p < 0.001; Fig. 2C). Therefore, we next determined whether any of these cells could be characterized as “place cells.” We compared spatial information scores with time-shuffled bootstrapped distributions (p > 0.01) and found that 43% of STZ group cells qualified, while only 29% of control units did (Fig. 2D). Representative examples of ACC place fields for both groups are depicted in Figure 2A. Taken together, these results revealed significant alterations in spatial coding in ACC units due to hyperglycemia.

Figure 2.

Biased spatial coding in ACC single units in hyperglycemia. A, Representative examples of spatially tuned cells. Place cells in ACC from healthy controls (top panel) and hyperglycemic animals (bottom panel). B, Violin plots for the mean firing rate (Hz) for ACC neurons in control versus STZ rats. C, Violin plots for normalized spatial information scores (Z) for ACC neurons in control versus STZ rats. Information scores were higher for hyperglycemic animals compared with their control counterparts. D, The percentage of significant spatial cells (place cells) versus nonspatial cells. E,F, Spatial information across the population of significant spatially tuned cells. ACC maximum spatial information plots for control (top) and STZ (bottom) plotted as color-coded rate maps, sorted by field positions on the maze, showing biased prereward coding in STZ animals compared with controls. G, Distribution of maximum spatial information location for significant spatially tuned cells. ACC place cells in hyperglycemic animals (pink) exhibited biased coding for prereward areas of the maze compared with controls, which were more evenly spread over the maze (blue).

After finding significant differences in single-unit coding properties of the ACC under hyperglycemia, we next examined if there were differences in spatial information by maze location (Fig. 2E,F). We calculated spatial information in each maze zone and then took the difference in maximum values for prereward zones (2–6) and postreward zones (8–12; Fig. 1A). For this analysis, positive values would reflect stronger prereward spatial coding and negative values signaled more postreward spatial information. As can be seen in Figure 2G, for control animals, the distribution was rather symmetrical, with a mean slightly >0. However, for hyperglycemic animals’ spatially tuned ACC cells, the distribution was heavily skewed with more positive values. We found that for control animals, the distribution was rather normal-shaped with a mean of 0.537, while hyperglycemic animal ACC neurons were more likely to have place fields before the reward with a mean of 1.21 pre–post reward ratio. This difference was significant in a one-way ANOVA (F_(1,215) = 7.112; p < 0.01; Fig. 2G), showing that hyperglycemic place cell spatial information was biased toward prereward maze locations.

ACC ensemble spatial reward coding was muted in hyperglycemic rats

Our analyses of ACC single units revealed increased spatial information and a higher percentage of place cells in hyperglycemic animals. We next investigated how the observed changes in spatial firing manifested across ACC ensemble coding. For population analyses, we included all ACC neurons, which included spatially tuned (place cells) and not spatially tuned cells. We concentrated on patterns of activity that were dominant across the entire population. First, we pooled all our neurons together for each condition (STZ or control), by extracting firing rates for the 12 maze positions for seven random correct trials for both left and right trips. We then formed 15 neuron ensembles (randomly selected on each draw) and repeated this process 1,000 times per area/condition group. We used PCA to extract the dominant patterns across the 15 neurons for the correct trials (7 left and 7 right). We then examined state separation in the top three PC space.

With our ACC ensemble analysis, we hoped to better understand how the hyperglycemic state was affecting ACC delayed alternation processing. Our analyses concentrated on differentiation between left and right trials and whether ensembles could be used to successfully decode the two trial types at the different locations on the maze.. In Figure 3A, an example ensemble showing clear low-dimensional trajectories over the course of the average left and right trials. To quantify the effects in state separation, we calculated the dMah in the top three PC space between left and right trials for the different maze locations. We found significant main effects for condition (F_(1,23976) = 10.973; p < 0.001) and maze location (F_(11,23976) = 192.74; p < 0.001; Fig. 3B). Follow-up tests revealed that ACC ensemble state separation was only different between groups at the reward port (p < 0.001) and in this case, control ensembles had significantly larger separation than in hyperglycemic animals.

Figure 3.

ACC ensemble spatial reward coding is muted in hyperglycemia. A, ACC state space trajectories during delayed alteration trials. Representative examples of low-dimensional trajectories in the state space separation in the top three PC space are plotted for ACC controls (left) and STZ ensembles (right). ACC ensembles did not show a difference for left/right selectivity across any maze location in both groups. Green arrows show the reward locations. B, State space separation was greater at reward location for control ensembles. State space separation between right and left trials was only different between the two groups at the reward location, where ACC controls (blue) had higher state space separation than ACC STZs (pink), indicative of diminished reward coding in hyperglycemic animals. C, Decoding accuracy by maze location. More accurate decoding of right versus left trials at reward location in controls. Using a linear decoder and leaving-one-out approach, decoding accuracies of ACC ensembles were comparable between the two groups over the entire maze, except at the reward location, where decoding accuracy was higher in controls (blue) compared with STZ (pink). D, State space separation matrix between all locations during right (y-axis) and left (x-axis) trials. Mean dMah between trial types and locations is plotted color. E, ACC control ensembles had a higher reward ratio than the STZ group. State space separation between right and left trials for all other locations compared with for reward locations. Stronger reward representation in ACC controls (blue) than in hyperglycemic animals (pink).

Using a linear decoder and a leave-one-out approach, we were able to successfully classify almost all locations for both conditions above chance levels. We found the highest left/right decoding accuracy (84.4%) at the reward location for control ensembles (Fig. 3C). When we examined separation between trials types at all locations, we found that reward location decoding was more accurate in control ensembles (Fig. 3D,E) compared with the STZ group. In Figure 3D we show similarity matrices based on the mean distance between left and right trials across all ensembles. The cross-like pattern at the reward location for control ensembles shows that two reward locations were not just distinct from each but also separated from other locations during the same trials. No such pattern appeared for hyperglycemic ACC ensembles. This indicates that the hyperglycemic state was affecting ACC ensembles’ spatial coding in different ways than we found in single cells, where we found increased spatial information (Fig. 2). To further elucidate the differences we observed at reward location for both ACC groups, we computed a reward ratio by dividing the mean reward location value by the mean separation for all other locations. We found strong evidence of reward overrepresentation in control ACC ensembles, and the effect was significantly smaller in STZ ensembles (F_(1,1998) = 175.18; p < 0.001; Fig. 3E). This can be clearly seen in Figure 3, B and C, where there is a clear increase in state space separation at the reward location that is not apparent in the mean of STZ ensembles. Overall, we found reward overrepresentation in healthy controls compared with STZ animals, indicating that hyperglycemia has diverse effects on ACC coding, especially with respect to reward location.

Hyperglycemia-linked to increased theta rhythmicity

There is a rich literature showing that neurons in the medial PFC, including the ACC, are strongly influenced by hippocampal theta activity during working memory (Jones and Wilson, 2005; Hyman et al., 2010; Hallock et al., 2016). Given that we have previously reported hyperglycemia driven changes in theta rhythm power in both the HPC and ACC, along with changes in coherence during working memory delays (Wirt et al., 2021), it was reasonable to investigate whether theta-related effects were behind the differences we observed in ACC ensembles. For each ACC neuron, we examined phase locking to both hippocampal and ACC theta rhythms during task performance. We then constructed time-shuffled empirical distributions for each neuron to determine whether the observed phase-locking effects were statistically significant. Overall, we found a slightly higher percentage of significantly theta phase-locked neurons in hyperglycemic animals were modulated by hippocampal theta (STZ, 45.6%; control, 38.3%) but less so for the population modulated by internal ACC theta (STZ, 17.9%; control, 28.2%). When we examined hippocampal theta phase-locking by position and group, we found significant main effects for both position (F_(10,6758) = 28.92; p < 0.001) and group (F_(1,6258) = 22.5; p < 0.001) and a significant interaction (F_(10,6258) = 3.23; p < 0.0001). Follow-up tests show that the hippocampal theta phase locking was stronger in the control group but only at the reward location (p < 0.001; Fig. 4A,B). For ACC theta phase-locking, we again found significant main effects for location (F_(10,6258) = 22.75; p < 0.001) and group (F_(1,6258) = 19.46; p < 0.001) and a significant interaction (F_(10,6258) = 1.98; p < 0.05). Follow-up tests found significant group differences only at Position 8, just after the reward area. Hyperglycemia leads to altered theta power in the HPC (Wirt et al., 2021), and here we report hyperglycemia leads to altered ACC neuron theta phase-locking for both hippocampal and ACC oscillations across the population, but not necessarily stronger phase locking for individual neurons, save for at the reward or reward adjacent locations.

Figure 4.

Hippocampal theta rhythmicity drives reward representation in ACC healthy controls. A, Theta phase locking by maze location. Mean hippocampal theta PLV on y-axis and maze location on the x-axis. Shading shows standard error of the mean (SEM). B, Mean ACC theta PLVs by maze location. C, Distribution of ACC units by theta phase locking. D, State space separation between trials by different subpopulations. Mean bootstrap ensemble dMah values are on the y-axis and maze location is on the x-axis. Plots show not theta (left), hippocampal theta phase-locked (middle), and ACC theta phase-locked ensembles (right). Control animal ensembles in solid lines and hyperglycemic neuron ensembles in dashed lines. p < 0.01. E, Reward location separation for all control animal bootstrap ensembles by subpopulations. Data show the dMah between right and left reward zones on the y-axis. Shading shows SEM. p < 0.01. F,G, Representative example peristimulus time histogram plots for control ACC neurons. Solid lines show mean response over all trials by condition, and shaded error bars show SEM. H,I, GLM results for control (H) and hyperglycemic (I) ACC neurons. Proportion of neurons with significant GLM results on the y-axis and maze location on the x-axis. J, Amount of significant neurons for task-phase GLM. Proportion of neurons with significant GLM results on the y-axis and maze location on the x-axis. K, Location selectivity for task GLM neurons. Mean left/right d′ values are on the y-axis and maze location on the x-axis. L, Location selectivity for location GLM neurons. Mean left/right d′ values are on the y-axis and maze location on the x-axis. M, Amount of significant neurons for task-phase and location GLMs. Proportion of neurons with significant GLM results for both models on the y-axis and maze location on the x-axis.

Decreased theta-linked reward coding in ACC

To understand how different subpopulations of ACC neurons were responding during the working memory task, we created bootstrap ensembles consisting of only hippocampal theta phase-locked, only ACC theta phase-locked, or not theta phase-locked neurons. We found that hippocampal theta phase-locked ensembles in control animals had significantly greater state space separation between right and left trials at multiple points on the maze than the hyperglycemic group (F_(1,23976) = 120.62). The effect was most pronounced at the reward location but was significant for locations after the reward as animals headed back into the delay box for the next trial (p < 0.0001; Fig. 4A). We found similar results in ACC theta phase-locked ensembles, where again control ensembles outperformed the hyperglycemic group at the reward location (F_(1,23976) = 126.43; p < 0.0001). Again, the greatest state space separation was at the reward location and was also significantly greater at the ensuing return arm locations (p < 0.00001; Fig. 4D). Interestingly, we also found significant differences between groups for the nontheta phase-locked ensembles overall (F_(1,23976) = 153.81; p < 0.0001), but follow-up tests found no differences at the reward location (Fig. 4D). For these ensembles constructed from nontheta phase-locked ACC neurons, we found that the hyperglycemic group had larger state space separation only at the end of the return arm as animals turned to enter the delay box (p < 0.0001). It should be noted that overall, we found comparable numbers of phase-locked neurons in both control and hyperglycemic groups, and this was true for both hippocampal and ACC theta rhythms (Fig. 4C). All together we found compelling evidence that theta phase-locked cells in hyperglycemic animals were encoding less information about reward locations than in control animals, likely leading to the muted ensemble reward overrepresentation.

Theta-modulated cells carried more reward information

For us to conclude that hyperglycemia was leading to altered reward representations in theta phase-locked cells, we would need to better understand how reward coding varied in control ensembles from theta phase-locked and not phase-locked neurons. We found that theta phase-locked (both HC and ACC theta) ACC ensembles had significantly larger reward location separation than nontheta phase-locked (F_(2,2997) = 69.52; p < 0.0001; Fig. 4E). While both hippocampal and ACC theta phase-locked ensembles had stronger reward coding than nontheta cells, hippocampal theta ensembles had significantly more reward separation than ACC theta ensembles (p < 0.0001). These results show that reward overrepresentation effects in ACC ensembles are driven by theta phase-locked cells and hyperglycemia leads to a specific deficit in reward information in theta phase-locked ACC cells.

Theta-modulated single units differentiated reward areas in control but not hyperglycemic animals

Next, we wanted to go back to single units to understand how ACC neurons responded over the course of delayed alternation trials. We used a series of GLMs to determine whether cell firing rates were significantly differentiating different trial epochs between left and right trials or “location models.” This allowed us to investigate how the ensemble results we found above tracked with single neuron effects. This approach also allowed us to look at how task epoch correlates (such as those apparent in the examples in Fig. 4F,G) varied between the subpopulations of neurons based on theta rhythmicity. After running the series of GLMs, we used Bonferroni’s correction to assess significance and then calculated the proportion of significant cells within each of our three subpopulations (HPC theta phase-locked, ACC theta phase-locked, and nontheta). As can be seen in Figure 4H, control animal HPC theta phase-locked displayed the highest percentage of significant cells during the reward approach and at the reward location. ACC theta phase-locked cells peaked just after the reward, during the return trip, and nontheta cells were equally spread throughout all trial epochs. In STZ animals (Fig. 4I), the distributions were largely flat for all subpopulations, with the largest percentage of significant cells appearing in the nontheta group after the reward during the return trip. The dynamics in the two theta phase-locked groups were highly divergent between control and STZ animals, with the biggest difference at the reward approach and reward locations (control, L6 = 25.3%; L7 = 30.1%; STZ, L6 = 9.9%; L7 = 14.3%). However, for ACC theta phase-locked cells, the biggest difference was at the reward and reward exit locations (control, L7 = 19.6%; L8 = 22.9%; L9 = 29.5%; STZ, L7 = 12.6%; L8 = 14.2%; L9 = 22.2%). For nontheta cells, the biggest differences between groups were at two return arm locations but in the opposite direction with more STZ cells being significant (control, L9 = 18.1%; L10 = 12.5%; STZ, L9 = 25.6%; L10 = 24.0%). These results support our ensemble analysis findings in two ways: (1) more significant firing rate changes at the reward approach and reward locations for both HPC and ACC theta phase-locked cells in control animals, and (2) these signals were notably absent in the chronically hyperglycemic animals. These analyses also revealed a tendency of STZ nontheta cells to be more active during the return trips where we also found these animals paused during their return trips (Fig. 1J). Perhaps this was evidence of some form of cognitive adaptation, as these cells were now relied on to carry this information that in healthy animals was present in theta-locked populations.

Reward expectation was intact, but reward location information was muted in hyperglycemic animals

Lastly, we were curious whether the differences we found in unit and ensemble coding were based on impaired reward location information or changes in reward expectation. We ran a second set of GLM models, but this time with equivalent activity for both left and right trials, we call these the “task models.” We found similar percentages of neurons were significant for task models in both groups over most of the maze. The biggest difference appeared at the decision point, but this was only a 7% difference, much smaller than found for numbers for the “location” GLM model. Next, we compared the left/right selectivity of the neurons that were significant for the “task” model. As can be seen in Figure 4K, the only significant difference between groups was at the reward location (p < 0.05), showing that reward expectation activity was spared in hyperglycemic animals but reward location information was muted.

To further examine whether the underlying factor was reward location or reward expectation, we looked again at the “location” model and examined the left/right selectivity of cells significant for this model. We reasoned that even though fewer hyperglycemic group cells significantly differentiated left and right trial locations, it was still possible that the cells that did had comparable left/right differential firing as the control group. We found that “location”-significant control group neurons carried more left/right information at the reward location only (p < 0.05; Fig. 4L). This further supports the idea that reward location and expectation are mixed in control animals, but reward location information was muted in hyperglycemic animals.

Lastly, we examined what the percentage of neurons was significant for both the “task” and “location” models. This would reveal whether reward location and expectation were intertwined or separated. We found that as the animals approached the reward location (L5 and 6; Fig. 4M), a large majority of reward expectation cells (i.e., “location” model significant) also coded strong reward location information (i.e., “task” model significant). Hyperglycemic animals were different, as less than half of neurons carried both reward location and expectation information until after the reward location had been passed (L8), with a peak at the last locations of the return arms, the same places we found hyperglycemic animals lingered more than controls before entering the sequester area (Fig. 1J). These analyses show that in hyperglycemic animal ACC neurons, reward expectation information was intact, but reward location was muted, unlike in controls where the two signals were largely mixed. All together, the single-cell explorations reveal some fundamental alterations in the neuronal activity across the ACC population that were driven by chronic hyperglycemia, but that did not affect behavioral accuracy.

Discussion

Chronic hyperglycemia, a major pathology of T2D, altered ACC information processing during a working memory task. Interestingly, behavior was largely unchanged, though hyperglycemic animals did not linger as long at the reward location after being rewarded. We found changes in ACC spatial processing centered around the reward zones, where there was a selective impairment in chronically hyperglycemic animals, in spite of an overall increased spatial information. Spatial coding allocation was biased in hyperglycemic ACC units, with a majority coding for prereward locations, in contrast to controls where spatial information was more evenly distributed throughout the maze. Ensemble analysis revealed higher state space separation and decoding accuracy for healthy controls but only at the reward and postreward locations. Lastly, hippocampal theta phase-locked ACC cells had stronger reward coding than ACC theta synchronized or nontheta synchronized ACC cells, indicating that reward representations in healthy controls were driven by hippocampal theta phase-locked cells. In chronically hyperglycemic animals on the other hand, theta phase-locked cells did not represent reward more strongly than nonphase-locked cells. Taken together, ACC information processing was affected at both the single unit and ensemble level, indicative of the heavy toll glucose dysregulation has on brain areas important for higher-order cognition, such as the ACC.

Hyperglycemia alters ACC unit activity in working memory tasks

Consistent with previous results, we found that in control animals, spatial information was distributed throughout the maze; however, chronically hyperglycemic animals had much stronger coding of prereward locations. This suggests that reward anticipation was heightened in these animals. Given T2D's clear links with obesity and known changes in sucrose preference (Yu et al., 2014), it is possible that ACC circuits are not receiving similar reward-related input from other brain areas such as the ventral striatum or hypothalamus. While hyperglycemic animals performed comparable with controls, the predisposition toward prereward information suggests the bias is related to anticipation of the reward. Such effects would be expected in trial discrete tasks, such as found in sequence tasks (Ma et al., 2014), but are less likely to be found in a continuous working memory task. Effects like this could appear if the ACC becomes more strongly modulated by striatal inputs in T2D, a possibility to examine in future experiments. It is also important to note that these differences were only apparent for single-cell analysis and ensemble-level dynamics revealed a different story. The disconnect between increased prereward location spatial information and no increase in ensemble left/right trial decoding for those locations suggests that hyperglycemic cells were more attuned to maze progression than to specific spatial locations. This was supported by comparisons of left/right differential firing in prereward areas, which was muted in the hyperglycemic group in spite of the increased spatial information, indicating that cells were likely coding reward expectation, not location. This would be consistent with results found in sequencing tasks, where ACC neurons fire at specific points in a series of behaviors (Hyman et al., 2010; Ma et al., 2014), or with time cells, where neurons fire at certain intervals within a specified temporal window (Ning et al., 2022). While such changes suggest that the animals may have been employing a different strategy, more detailed behavioral experiments are needed. Perhaps a task requiring pure spatial alternation as opposed to the response alternation that can be used in the figure 8 maze may uncover hyperglycemia-linked behavioral deficits.

Hyperglycemia leads to muted signaling and altered behavior at the reward location

The uneven results of ACC single-unit analysis led us to gauge the effect of hyperglycemia on ACC network activity. State space separation and decoding accuracy analysis only found separation differences over left and right trials between the two groups at the reward location, which was significantly greater in controls. In contrast, this reward representation was “muted” by chronic hyperglycemia, perhaps indicative of the effect of altered metabolic state on ACC population reward coding. Multiple studies on ACC acknowledge its central role in controlling the diverse aspects of reward-guided behaviors, such as identifying the optimal environmental stimulus and relating actions with outcomes (Bush et al., 2002; Rushworth et al., 2004; Walton et al., 2004; Hyman et al., 2013). It is also involved in assessing reward proximity (Shidara and Richmond, 2002) and reward-based sequence performance and learning (Procyk et al., 2000). The ACC exerts top–down control on other prefrontal areas (Kerns et al., 2004), eventually leading to optimal goal-directed behavior. The current ACC effects are an ideal target to understand how metabolic disorders affect the coding of reward and how the ACC might affect such processes. Given that lifestyle interventions are often a key component of T2D treatment and are notoriously challenging to maintain, with only ∼20–60% of patients adhering (Ganiyu et al., 2013; Mumu et al., 2014), understanding how reward-related information is altered by T2D pathology could help shape future lifestyle programs.

We found that hyperglycemic animals spent less time at the reward location on correct trials. While controls exhibited a typical postreinforcement pause, hyperglycemic animals quickly moved on from the reward site. This could be indicative of anhedonia, a characteristic often seen in depressed patients, and that is closely associated with T2D and vascular disease, potentially explaining the increased co-morbidity of T2D and depression (Carter and Swardfager, 2016; Willame et al., 2022). We should also note that in the current report, while we found decreased reward zone lingering, we only found one other location with behavioral differences, while ACC neurons responded differently throughout the maze. We would argue that the decreased reward location coding seen both leading to and at the reward site are consistent with hyperglycemic animals not experiencing reward the same as controls did, which is further supported by their lack of postreinforcement pause. Additionally, the longer time hyperglycemic animals spent in the return arms perhaps represents a more concerted effort by hyperglycemic animals to form a future retrieval cue for the next trial since they rushed away from the reward zone where such cues are likely normally formed. Further research into the role of anhedonia, T2D, depression, and glycemic dysfunction should take into account the altered information coding we found between the HPC and ACC.

Possible cellular mechanisms for altered ACC unit activity in hyperglycemic animals

STZ, a toxin, has an analogous molecular structure to d-glucose and can therefore enter cells via ordinary GLUT-2–mediated process and selectively influence GLUT-2 containing GM (glucose-monitoring) cells. A recent study found these GM neurons in ACC (Hormay et al., 2019), which were found to be selectively destroyed by STZ injections, causing various feeding-metabolic alterations. The ACC is also known to be reciprocally connected to several brain areas such as prefrontal and insular cortices, the amygdala, nucleus accumbens, striatum, and hypothalamus, all of which are known to be a part of the central GM neural system. Therefore, ACC is an important brain region for homeostatic regulation of the metabolic as well as the reward system. Glucose dysregulation may consequently lead to modified ACC reward information processing, perhaps by altered perception of reward as a salient feature and motivator, consequently reducing a reward's abstract value (Yu et al., 2010; Koekkoek et al., 2017; Alvarsson and Stanley, 2018).

Another possible mechanism is myo-inositol, a brain metabolite found in ACC, which is responsible for cellular osmotic balance, membrane phospholipid turnover, and insulin-modulated brain network functional connectivity. Myo-inositol is elevated in diabetes, MCI, and even AD and has been linked to altered functional connectivity. Elevated plasma insulin levels decrease pregenual ACC myo-inositol levels and increase functional connectivity between sensorimotor regions and ACC-/insula-related networks (Bolo et al., 2020). However, in hyperglycemia and diabetes, due to decreased insulin levels, there is an increase in ACC myo-inositol levels, leading to network dysfunction and related cognitive impairments we observed in this study. The destruction of GM neurons and increased levels of myo-inositol in ACC during hyperglycemia may consequently be responsible for decreased reward coding in STZ animals compared with healthy controls. Further research is needed to understand the mechanisms driving the muted reward coding we observed here, and myo-inositol and GM cells are two avenues to pursue.

Reward location information stronger in theta-modulated ACC neurons

We found that reward location information was strongest in theta phase-locked ACC neurons in control animals, but in hyperglycemic animals, there were no differences in reward information. This indicates that reward information was selectively muted in these theta-responsive cells, suggesting a breakdown in processing that likely affected the entire limbic system. This would be consistent with previous work from our lab showing theta alterations in the same T2D model and also findings revealing the strong effects insulin receptors have on HPC function. Alterations of the insulin receptor gene in mice were shown to increase food intake that then leads to obesity which in turn increases the risk of developing diabetes and hyperglycemia (Brüning et al., 2000). In the mouse, insulin can either enhance or inhibit local field excitatory synaptic postsynaptic potentials depending on internal circumstances promoting LTP and modulating LTD induction, both important to the learning and memory process (Zhao et al., 2019) and both highly dependent on theta rhythm (Hyman et al., 2003). Muted reward information in theta-responsive cells may be attributed to changes in insulin availability which could lead to conformational changes to insulin receptors that over time disrupts LTP and LTD. Furthermore, dysfunction in the ACC-HPC circuit has been linked with the transition from healthy aging to MCI and is an area of interest for AD progression and treatment (Calvin-Dunn et al., 2025).

Another important aspect of these results is the implication that in healthy brains, reward location information is carried by theta frequency communication. Selective communication between different brain areas that are timely and synchronized via cell assemblies is key to neural processing and information transfer. Many important cognitive tasks involving spatial learning and memory, decision-making, and reward encoding are contingent on prefrontal and limbic areas of the brain, including the HPC and ACC. Hippocampal theta synchronizes to mPFC (Jones and Wilson, 2005; Siapas et al., 2005) and ACC (Hyman et al., 2005) depending on a task's cognitive demands, ultimately leading to updated maintenance of working memory (Hyman et al., 2010; Hallock et al., 2016; Zielinski et al., 2019). HPC–PFC theta synchrony is known to increase when determining reward probability at the choice port (Benchenane et al., 2010), during the delay period in a working memory task (Myroshnychenko et al., 2017), and at decision stage in a trial (Jones and Wilson, 2005). The HPC is known to precede the cingulate in task-relevant information processing which is later on transferred to cingulate population via theta coherence (Remondes and Wilson, 2013). A recent study revealed that population coding in ACC encodes spatial location more accurately than single-unit coding, and this accuracy increases with reward- and goal-directed navigation (Ma et al., 2023). The current results show that reward information is a product of multiple neural areas, with different features about the reward being coded by different areas, notably here, reward location in the HPC.

Our results revealed that spatial coding mechanisms are impaired and affected differentially at both single unit and ensemble level in the ACC, indicative of the complex and layered effects of glucose dysregulation on the brain. Future studies focused on spatial coding aspects of the PFC and other higher-order brain areas are needed to better understand the link between T2D and cognitive impairments, which can help inform why T2D is such a powerful risk factor for neurodegenerative disorders such as AD. These studies are imperative in determining higher-quality therapeutic interventions that may be targeted at specific higher-order brain areas such as the ACC to combat such devastating neurodegenerative diseases in humans.